General comments
From my experience and after a looking in the amp. schematic, i have two remarks. The first it is that, this amp. has a CFP type output stage. The other type - and most commonly used - it is the EF output. In comparisson between them, the CFP appeared as some better in frequency response but it is also most delicate from the EF which is by far most rugged. Also in class AB the CFP output produces more heat than EF in iddle state thus it needs bigger heatsinks. A concrect drawback of CFP it is that, if the output transistor blocked i.e. under a saturation condition, then all the output current pass through the smallest driver to the load and then it destroyed. You can make as i think an easy experiment to convert the output in EF. Leave as they are the drivers, move the 220Ù resistors between the emitters of BDs and the output rail, connect the collectors of BDs in supply rails, interchange the output transistors between them (3055 in +supply rail and 2955 in -rail) disconnect the base of each output transistor from the collector of each driver and reconnect it in the emiter of the driver. Leave the collector of each output transistor connected in supply rails directly. Thus you can obtain a EF arrangement in output. Try it and make a comparison in ruggedness and the sound quality. One more remark, it is the absence at least of the output clamping diodes which protect the output devices from the unwanted returned currents from the load to the output transistors during the opposite cycle of output swing, which can apply a peak voltage of 70V (in your case with the +/-35V supply level) accross the c-e contacts of the inactive transistor during this momment and don't forget that although 3055-2955 has a Ic=16A their Vce it is only 60V. Place two 1N4004 (for certainity) each one accross the output rail and each one of the supply rails in a reverse connection. Be carefull to include also the emitter resistors into!
I think you resolve your problem with this modification.
Fotios